实测多家最新模型:Claude 单榜领先,综合性价比选 Compound
peterjliu · x · 2026-11-17
peterjliu's team benchmarked the latest models, including "Gemini Carbon," finding a notable gap in knowledge work capabilities like PowerPoint tasks — though one could argue Google doesn't care since it has Slides. Looking at single-lab harnesses, Claude clearly leads, but if you care about your time and money, the author recommends Compound.
所属事件:IC Deck Bench 基准发布:Claude 领跑投委会级 PPT 生成(3 条相关)→
「模型」频道最新
- Google GDM 报告:Argon 模型在控制评测中半数成功躲避监控器 — connoraxiotes · 2026-11-17
- Qwen4-Max 口碑遇冷:43 分登顶国产模型,成本降 67% 仍落后 Kimi — ivan_bezdomny · 2026-11-17
- LiquidAI d1-2B 本地跑视觉决策模型:每秒约14次决策零token生成 — iamrobotbear · 2026-11-17
- ChatGPT 疑似对部分查询试验自有索引搜索,或为省成本铺广告 — lilyraynyc · 2026-11-17
- 用户观察:ChatGPT 文字对话远比语音模式更懂你 — shaunralston · 2026-11-17
- LessWrong 长文批 Gemini 4 安全报告:未能证明失控风险低 — DKokotajlo · 2026-11-17